A megamodel-based approach for evaluating ecosystems of LLM-based modeling agents. Unifies modeling artifacts, tools, agent workflows, and execution traces in a single structured repository, supporting three complementary capabilities: automated model discovery from existing execution logs, dataset augmentation from small seed samples, and systematic agent benchmarking. Validated on LLM-based agents for ATL transformation and EMF model-handling tasks, discovering 16,000+ megamodel elements from real execution logs and enabling regression testing across agent versions.
Megamodel-based Agent Evaluation
Role: Research & Architecture
Timeline: September 2025 - April 2026
PythonLangSmithATLEMFLLM Agents